US baby names dataset

This notebook gives an example of Active Anomaly Detection with coniferest and US baby names dataset.

Developers of conferest: - Matwey Kornilov (MSU) - Vladimir Korolev - Konstantin Malanchev (LINCC Frameworks / CMU), notebook author

The tutorial is co-authored by Etienne Russeil (LPC)

Run this NB in Google Colab

Install and import modules

%pip install coniferest
%pip install pandas
%pip install requests

Requirement already satisfied: coniferest in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (0.0.15)
Requirement already satisfied: click in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from coniferest) (8.1.7)
Requirement already satisfied: joblib in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from coniferest) (1.4.2)
Requirement already satisfied: numpy in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from coniferest) (2.2.0)
Requirement already satisfied: scikit-learn<2,>=1.4 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from coniferest) (1.5.2)
Requirement already satisfied: matplotlib in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from coniferest) (3.9.3)
Requirement already satisfied: onnxconverter-common in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from coniferest) (1.14.0)
Requirement already satisfied: scipy>=1.6.0 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from scikit-learn<2,>=1.4->coniferest) (1.14.1)
Requirement already satisfied: threadpoolctl>=3.1.0 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from scikit-learn<2,>=1.4->coniferest) (3.5.0)
Requirement already satisfied: contourpy>=1.0.1 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (1.3.1)
Requirement already satisfied: cycler>=0.10 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (0.12.1)
Requirement already satisfied: fonttools>=4.22.0 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (4.55.2)
Requirement already satisfied: kiwisolver>=1.3.1 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (1.4.7)
Requirement already satisfied: packaging>=20.0 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (24.2)
Requirement already satisfied: pillow>=8 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (11.0.0)
Requirement already satisfied: pyparsing>=2.3.1 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (3.2.0)
Requirement already satisfied: python-dateutil>=2.7 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from matplotlib->coniferest) (2.9.0.post0)
Requirement already satisfied: onnx in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from onnxconverter-common->coniferest) (1.17.0)
Requirement already satisfied: protobuf==3.20.2 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from onnxconverter-common->coniferest) (3.20.2)
Requirement already satisfied: six>=1.5 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from python-dateutil>=2.7->matplotlib->coniferest) (1.17.0)
Note: you may need to restart the kernel to use updated packages.
Requirement already satisfied: pandas in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (2.2.3)
Requirement already satisfied: numpy>=1.23.2 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from pandas) (2.2.0)
Requirement already satisfied: python-dateutil>=2.8.2 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from pandas) (2.9.0.post0)
Requirement already satisfied: pytz>=2020.1 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from pandas) (2024.2)
Requirement already satisfied: tzdata>=2022.7 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from pandas) (2024.2)
Requirement already satisfied: six>=1.5 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from python-dateutil>=2.8.2->pandas) (1.17.0)
Note: you may need to restart the kernel to use updated packages.
Requirement already satisfied: requests in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (2.32.3)
Requirement already satisfied: charset-normalizer<4,>=2 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from requests) (3.4.0)
Requirement already satisfied: idna<4,>=2.5 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from requests) (3.10)
Requirement already satisfied: urllib3<3,>=1.21.1 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from requests) (2.2.3)
Requirement already satisfied: certifi>=2017.4.17 in /home/docs/checkouts/readthedocs.org/user_builds/coniferest/envs/latest/lib/python3.11/site-packages (from requests) (2024.8.30)
Note: you may need to restart the kernel to use updated packages.

import zipfile
from io import BytesIO

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import requests
from coniferest.pineforest import PineForest
from coniferest.isoforest import IsolationForest
from coniferest.pineforest import PineForest
from coniferest.session import Session
from coniferest.session.callback import TerminateAfter, prompt_decision_callback, Label

Data preparation

Download data and put into a single data frame

%%time

URL = 'https://www.ssa.gov/OACT/babynames/state/namesbystate.zip'

response = requests.get(URL)
if not response.ok:
    raise RuntimeError('Cannot download the file')

dfs = []
with zipfile.ZipFile(BytesIO(response.content)) as zip_ref:
    for filename in zip_ref.namelist():
        if not filename.lower().endswith('.txt'):
            continue
        with zip_ref.open(filename) as f:
            df = pd.read_csv(f, header=None, names=['State', 'Gender', 'Year', 'Name', 'Count'])
        dfs.append(df)
raw = pd.concat(dfs, axis=0, ignore_index=True)

raw

CPU times: user 3.02 s, sys: 477 ms, total: 3.5 s
Wall time: 3.92 s

	State	Gender	Year	Name	Count
0	AK	F	1910	Mary	14
1	AK	F	1910	Annie	12
2	AK	F	1910	Anna	10
3	AK	F	1910	Margaret	8
4	AK	F	1910	Helen	7
...	...	...	...	...	...
6504156	WY	M	2023	Parker	5
6504157	WY	M	2023	Rhett	5
6504158	WY	M	2023	Roman	5
6504159	WY	M	2023	Ryan	5
6504160	WY	M	2023	Timothy	5

6504161 rows × 5 columns

Let’s load the data and transform it into a feature matrix where each column is the number (normalized by peak value) of US citizens that got this name in a given year. We apply few quality filter. We require the name to appear at least 10000 times over the full time range, this will prevent noisy data from names that are barely used.

Optionally, we use first few Fourier terms to better detect “bumps” and “waves” in the time-series.

%%time

WITH_FFT = True

all_years = np.unique(raw['Year'])
all_names = np.unique(raw['Name'])

# Accumulate names over states and genders
counts = raw.groupby(['Name', 'Year']).apply(lambda df: df['Count'].sum())

# Tranform to dataframe where names are labels, years are columns and counts are values (features)
years = [f'year_{i}' for i in all_years]
year_columns = pd.DataFrame(data=0.0, index=all_names, columns=years)
for name, year in counts.index:
    year_columns.loc[name, f'year_{year}'] = counts.loc[name, year]

# Account for total population changes
trend = year_columns.sum(axis=0)
detrended = year_columns / trend

# Normalise and filter
norm = detrended.apply(lambda column: column / detrended.max(axis=1))
filtered = norm[year_columns.sum(axis=1) >= 10_000]

if WITH_FFT:
    # Fourier-transform, normalize by zero frequency and get power-spectrum for few lowest frequencies
    power_spectrum = np.square(np.abs(np.fft.fft(filtered)))
    power_spectrum_norm = power_spectrum / power_spectrum[:, 0, None]
    power_spectrum_low_freq = power_spectrum_norm[:, 1:21]
    frequencies = [f'freq_{i}' for i in range(power_spectrum_low_freq.shape[1])]
    power = pd.DataFrame(data=power_spectrum_low_freq, index=filtered.index, columns=frequencies)

    # Concatenate time-series data and power spectrum
    final = pd.merge(filtered, power, left_index=True, right_index=True)
else:
    # Use time-series data
    final = filtered

final

<timed exec>:7: DeprecationWarning: DataFrameGroupBy.apply operated on the grouping columns. This behavior is deprecated, and in a future version of pandas the grouping columns will be excluded from the operation. Either pass `include_groups=False` to exclude the groupings or explicitly select the grouping columns after groupby to silence this warning.

CPU times: user 1min 53s, sys: 459 ms, total: 1min 53s
Wall time: 1min 53s

	year_1910	year_1911	year_1912	year_1913	year_1914	year_1915	year_1916	year_1917	year_1918	year_1919	...	freq_10	freq_11	freq_12	freq_13	freq_14	freq_15	freq_16	freq_17	freq_18	freq_19
Aaliyah	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	...	0.006223	0.011881	0.012348	0.005589	0.000240	0.001929	0.005152	0.004159	0.000908	0.000148
Aaron	0.045718	0.053744	0.062024	0.077323	0.074977	0.065559	0.066898	0.066008	0.065487	0.066251	...	0.000102	0.000375	0.000008	0.000174	0.000008	0.000116	0.000323	0.000183	0.000018	0.000206
Abbey	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	...	0.003382	0.002492	0.000490	0.000341	0.000792	0.000629	0.000184	0.000041	0.000019	0.000142
Abbie	0.359932	0.269787	0.306439	0.464589	0.215530	0.302400	0.349221	0.307588	0.197794	0.346314	...	0.000872	0.000340	0.000209	0.000339	0.001921	0.000232	0.000076	0.000102	0.000040	0.000605
Abbigail	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	...	0.002642	0.001734	0.001141	0.001235	0.001421	0.001037	0.000383	0.000060	0.000073	0.000134
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
Zelma	0.932474	1.000000	0.729411	0.723820	0.652394	0.669620	0.600107	0.576397	0.536196	0.576366	...	0.012419	0.010827	0.009236	0.007330	0.005842	0.005533	0.005881	0.005967	0.006053	0.006224
Zion	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	...	0.029102	0.015499	0.004003	0.004927	0.011328	0.011587	0.005323	0.001245	0.003695	0.007141
Zoe	0.009243	0.000000	0.003225	0.000000	0.001845	0.002542	0.004274	0.004372	0.002133	0.001221	...	0.007844	0.009167	0.006538	0.002683	0.000962	0.000918	0.001534	0.002198	0.002286	0.002146
Zoey	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	...	0.009403	0.011478	0.011382	0.009234	0.005897	0.002628	0.000721	0.000420	0.000919	0.001503
Zuri	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	...	0.049637	0.039396	0.035679	0.034023	0.030917	0.027702	0.022838	0.016293	0.011780	0.010174

2309 rows × 134 columns

Plotting function

def basic_plot(idx):
    cols = [col.startswith('year') for col in final.columns]
    all_years = [int(col.removeprefix('year_')) for col in final.columns[cols]]

    if isinstance(idx, str):
        counts = final.loc[idx][cols]
        title = idx
    else:
        counts = final.iloc[idx][cols]
        title = final.iloc[idx].name

    plt.plot(all_years, counts.values)
    # plt.ylim(-0.1, 1.1)
    plt.title(title)
    plt.xlabel('Year')
    plt.ylabel('Count')
    plt.show()

We can now easily look at the evolution of a given name over the years

print(len(final))
basic_plot('Anastasia')
basic_plot('Leo')

2309

Classical anomaly detection

model = IsolationForest(random_seed=1, n_trees=1000)
model.fit(np.array(final))
scores = model.score_samples(np.array(final))
ordered_scores, ordered_index = zip(*sorted(zip(scores, final.index)))

print(f"Top 10 weirdest names : {ordered_index[:10]}")
print(f"Top 10 most regular names : {ordered_index[-10:]}")

Top 10 weirdest names : ('Manuel', 'Alfonso', 'Marshall', 'Vincent', 'Margarita', 'Byron', 'Ignacio', 'Benito', 'Anthony', 'Ruben')
Top 10 most regular names : ('Aleah', 'Kenzie', 'Camden', 'Jayla', 'Haven', 'Kian', 'Bryson', 'Teagan', 'Holden', 'Raegan')

Let’s have a look at their distributions

for normal in ordered_index[-4:]:
    basic_plot(normal)

for weird in ordered_index[:4]:
    basic_plot(weird)

It seems that anomalies are either very localised peak or very recent trending names.

Active Anomaly Detection

First, we need a function helping us to make a decision.

Comment dummy decision function and uncomment interactive one

# Comment
def help_decision(metadata, data, session):
    """Dummy, says YES to everything"""
    return Label.ANOMALY

### UNCOMMENT
# def help_decision(metadata, data, session):
#     """Plots data and asks expert interactively"""
#     basic_plot(metadata)
#     return prompt_decision_callback(metadata, data, session)

Let’s create a model and run a session

Let’s run PineForest and say YES every time we see a recent growth

model = PineForest(
    # Number of trees to use for predictions
    n_trees=256,
    # Number of new tree to grow for each decision
    n_spare_trees=768,
    # Fix random seed for reproducibility
    random_seed=0,
)
session = Session(
    data=final,
    metadata=final.index,
    model=model,
    decision_callback=help_decision,
    on_decision_callbacks=[
        TerminateAfter(10),
    ],
)
session.run()

<coniferest.session.Session at 0x7f361c98f9d0>

Wow! Good almost 100% of the behavior we were looking for

Now we can run it again and say YES every time we see a sharp peak

model = PineForest(
    # Number of trees to use for predictions
    n_trees=256,
    # Number of new tree to grow for each decision
    n_spare_trees=768,
    # Fix random seed for reproducibility
    random_seed=0,
)
session = Session(
    data=final,
    metadata=final.index,
    model=model,
    decision_callback=help_decision,
    on_decision_callbacks=[
        TerminateAfter(20),
    ],
)
session.run()

<coniferest.session.Session at 0x7f35fc540a50>

PineForest learns the profile interesting for the user and outputs it.

Try to change Pineforest hyperparameters and see how results change. Try different random seeds and number of trees.